Return-path: Received: from mga02.intel.com ([134.134.136.20]:38606 "EHLO mga02.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753378Ab0CZDhG (ORCPT ); Thu, 25 Mar 2010 23:37:06 -0400 Subject: Re: [PATCH 1/2] mac80211: support paged rx SKBs From: Zhu Yi To: Stanislaw Gruszka Cc: Johannes Berg , "linville@tuxdriver.com" , "linux-wireless@vger.kernel.org" In-Reply-To: <20100325072627.GA2562@dhcp-lab-161.englab.brq.redhat.com> References: <1269399463-5244-1-git-send-email-yi.zhu@intel.com> <20100324140034.6a6d91c9@dhcp-lab-109.englab.brq.redhat.com> <1269447362.6935.2.camel@jlt3.sipsolutions.net> <20100325072627.GA2562@dhcp-lab-161.englab.brq.redhat.com> Content-Type: text/plain; charset="UTF-8" Date: Fri, 26 Mar 2010 11:37:51 +0800 Message-ID: <1269574671.4043.189.camel@debian> Mime-Version: 1.0 Sender: linux-wireless-owner@vger.kernel.org List-ID: On Thu, 2010-03-25 at 15:26 +0800, Stanislaw Gruszka wrote: > > But wait, isn't that all just workaround for bad performing allocator, > what should be fixed in proper place not in mac80211 nor iwlwifi. What > for exactly this is needed, concretes please - that is not described > in patch changelog. If a devices is capable of DMA a frame to/from physically discontinuous host memory, the paged SKB handling can mitigate memory subsystem pressure for large chunk of continuous memory allocation and also avoid CPU cycles for memcpy. In Linux kernel, since TCP/IP stacks and (capable) devices support paged SKBs, mac80211 will become the bottleneck if it still couldn't provide the support. For devices don't support paged SKBs, this patch doesn't bring performance issues. So it should be a good thing to have for us. Thanks, -yi